Molecular characterization of acquired resistance to KRAS G12C inhibition in gastrointestinal cancers

KRAS G12C inhibitors, such as sotorasib, have rapidly moved through clinical development and are poised to transform care of patients with KRAS G12C mutant cancers, in particular non-small cell lung cancer (NSCLC) and colorectal cancer (CRC). Clinical e�cacy is achieved in NSCLC as a single agent and in CRC in combination with anti-EGFR monoclonal antibodies, however, secondary resistance impairs the effects of KRAS G12C blockade. In this work, we sought to determine the mechanisms of acquired resistance to concomitant KRAS-EGFR inhibition. In cell lines, patient-derived xenograft, and patient samples, a heterogeneous pattern of putative resistance alterations expected primarily to prevent inhibition of ERK signalling by drug can be detected at progression. Serial analysis of patient blood samples on treatment demonstrates that most of these alterations are detected at a low frequency that does not increase substantially and sometimes disappears over time, with the exception of KRAS G12C ampli�cation which rises in step with tumour marker levels and clinical progression. Here we show that a CRC cell line that acquired resistance to sotorasib-cetuximab combination through KRAS G12C ampli�cation became addicted to these agents and undergoes oncogene-induced senescence upon drug withdrawal. Accordingly, the KRAS G12C signal in circulating DNA from relapsed patients harbouring G12C ampli�cation rapidly recedes upon treatment holiday. These data indicate that KRAS G12C ampli�cation is a recurrent resistance mechanism to KRAS-EGFR co-inhibition and suggest a potential therapeutic vulnerability, whereby therapies that target this senescence response at drug withdrawal may overcome resistance to KRAS G12C-EGFR inhibition.


Full Text
KRAS is the most commonly mutated oncogene in human cancer 1 .It acts as a signalling switch that, when bound to GTP, orchestrates a program of cell proliferation and survival.Until recently, efforts to target KRAS have been unsuccessful due to its small binding pocket, high a nity for GTP, and redundant mechanisms of posttranslational processing.The development of allele speci c KRAS G12C inhibitors that trap KRAS in the inactive, GDP-bound state 2,3 led to a paradigm change, with clinical responses in 30-50% of non-small cell lung cancer (NSCLC) patients harbouring KRAS G12C mutations 4,5 .These agents are not as effective in colorectal cancers (CRC) with KRAS G12C mutation.We have previously shown that the activity of these drugs in KRAS G12C CRC is limited because activation of epidermal growth factor receptor (EGFR) reactivates ERK signalling and consequently combinatorial KRAS G12C and EGFR inhibition more effectively targets KRAS G12C CRC 6 .Early trial data provide clinical support for this observation: the response rate for sotorasib was 7-10% in CRC 7 and, in the rst report of sotorasib plus the EGFR antibody panitumumab, the response rate was 27% 8 .For adagrasib monotherapy, it was ~20% and, for adagrasib with the EGFR antibody cetuximab, it increased to ~40% 9 .Based on these data, these KRAS G12C and EGFR antibody combination treatments are being evaluated in registrational, phase 3 trials.
Nonetheless, patients treated with these agents eventually acquire resistance and the response to single agent or combination treatment is brief.Several studies have characterized resistance to KRAS G12C monotherapy [10][11][12] .Remarkably, these alterations are highly heterogeneous, including KRAS, BRAF, or MEK mutations, as well as gene ampli cations and fusions, and circulating tumour DNA (ctDNA) analysis typically identi es multiple resistance alterations in the same patient.
To identify mechanisms of resistance to the combination of KRAS G12C and EGFR inhibitors, we grew the CRC cell lines C106 and RW7213, both of which are sensitive to this treatment 6 in drugs until the emergence of secondary resistance (Figure 1a).Treatment with sotorasib 3mM and cetuximab 50mg/mL led to massive cell death of both cell lines with few viable cells.Cells were therefore subjected to increasing doses of sotorasib (from 0.1mM to 3mM) with cetuximab 50mg/mL to generate resistance (Extended Data Figure 1a).Resistant sublines grew well in drugs after a period of 4 months for C106 cells and 2 months for RW7213 cells (Extended Data Figure 1a, 1b, 1d).
Both resistant sublines expressed higher RAS-GTP levels than parental cells and drug treatment led to incomplete inhibition of RAS-GTP (Extended Data Figures 1c, 1e).Drug treatment of C106 cells failed to suppress activity of downstream effectors in the RAS/ERK pathway, whereas RW7213 resistant cells experienced a reduction in pathway activity but continued to have high levels of phospho-MEK and phospho-ERK due to elevated baseline pathway activation.Targeted sequencing of the resistant sublines using MSK-IMPACT 13 identi ed acquired NRAS G12D mutation in C106 cells and KRAS G12C ampli cation in RW7213 cells (Extended Data Figure Table 1).Single cell sequencing of the C106 resistant subline indicated that the NRAS G12D mutation occurred in the same cells bearing the KRAS G12C alteration (Figure 1b).The variant allelic frequency (VAF) of the NRAS mutation suggested allelic imbalance with an acquired gain of the NRAS allele, which was con rmed by bulk targeted exome sequencing (MSK-IMPACT).This seems to be a late event occurring in a subpopulation of the cells.More than 20 copies of KRAS per cell were detected in resistant RW7213 cells with uorescence in situ hybridization (Figure 1c).In parallel, a KRAS G12C mutant CRC patient-derived xenograft (PDX) model (CLR113) that was initially sensitive to sotorasib and cetuximab combination treatment 6 (Extended Data Figure 1f) developed acquired resistance after about 10 months that was associated with KRAS G12C ampli cation, BRAF K601E, and RAF1 S259F acquired alterations (Figure 1d).These data indicate that multiple resistance mechanisms can contribute to the survival of KRAS G12C mutant cells and that KRAS G12C ampli cation is a recurrent alteration at resistance.
To evaluate potential resistance mechanisms to KRAS G12C inhibitor (KRAS G12Ci) and anti-EGFR antibody (EGFRi) in patients, we collected circulating free DNA (cfDNA) from eleven CRC patients treated with combination treatment who initially experienced tumour regression and then developed either radiographic (RECIST) or clinical progression (Extended Data Figure 2 and Extended Data Table 2).Emergent alterations identi ed at resistance (Figure 1e) included KRAS G12C ampli cation, KRAS mutations (G12A/D/F/LR/S/V, H95L/Q/R, and Y96D/H/N), NRAS mutation (Q61K/R), downstream ERK pathway alterations (BRAF mutations/fusions, MEK1 mutations), RTK activation (MET ampli cation/fusion, RET fusion, EGFR mutations), and MYC ampli cation.Similar to what was previously reported for resistance to KRAS G12C inhibitor monotherapy and in accordance with our preclinical models treated with sotorasib-cetuximab combination, multiple resistance-associated alterations were identi ed in individual patients, with the majority predicted to prevent inhibition of ERK signalling by drug 12 .
Similar to resistance alterations to KRAS G12C inhibitor monotherapy 10,11 , all acquired alterations were identi ed at low VAF, at one-tenth or one-hundredth of the frequency of alterations identi ed at baseline.
To better understand clonal dynamics of resistance in gastrointestinal cancers, cfDNA was serially collected during treatment and sequenced about every six weeks in four CRC, one pancreatic cancer, and one small bowel cancer patients (Figure 2, Extended Data Figure 3, and Extended Data Table 3).
Longitudinal analysis con rmed the emergence of multiple resistance alterations.Resistant alterations often emerged many weeks before the development of clinical resistance and remained at a low frequency, largely <1% VAF, while the baseline alterations and tumour marker (CEA for CRC and small bowel cancer patients or Ca19-9 for pancreatic cancer patient) rose higher during treatment.In multiple patients, once resistance was rst detected, each successive time point identi ed new resistance alterations with only modest changes in the VAF of the pre-existing resistant alterations.We did not observe a clonal sweep with emergence of a dominant resistance alteration in any patient.In several patients, the VAF of the putative resistance alterations actually decreased and became undetectable despite continued treatment.These included alterations expected to cause resistance to a KRAS G12C inhibitor, such as the KRAS G13D, NRAS Q61K, and BRAF V600E mutations detected in patient 1 (Figure 2a) and the BRAF V600E mutation and BRAF fusion in patient 3 (Figure 2c).These data together suggest that resistant subclones do not grow effectively and are unable to grow out to dominate the population.Indeed, among the many low frequency resistance alterations detected in patients, the only putative resistance genetic event that correlated with tumour marker response was KRAS G12C ampli cation.Clinical resistance to KRAS G12C inhibition is thus characterized by the accumulation and loss of many low frequency resistance alterations, while KRAS G12C ampli cation drives a higher portion of the resistance phenotype.
Intrigued by the correlation of KRAS G12C ampli cation with clinical resistance, we used the RW7213resistant cells (RW7213-R) harbouring high-grade KRAS G12C ampli cation to investigate the characteristics of this resistance mechanism.The resistant cells grow well as colonies in medium containing cetuximab and sotorasib, maintaining the same morphology of the parental RW7213 cell line.However, these cells failed to proliferate, acquired a large and at morphology, and accumulated vacuoles 5 to 7 days after drug withdrawal (Figure 3a) suggesting that RW7213-R cells became addicted to the drug combination.The morphological features of these cells are reminiscent of cellular senescence, a tumour suppressive program that can be triggered by excessive oncogenic signaling 14 .In addition to acquiring a proliferative arrest and protection from apoptosis, senescent cells activate a secretory program known as the senescence associated secretory phenotype (SASP) 15,16 .Also in line with a senescent phenotye, RW7213-R cells taken off the drug combination downregulated apoptosis, increased expression of cyclin-dependent kinase inhibitors and uPAR, a marker of senescence 17 , and accumulate cytokines indicative of SASP (Figure 3b, 3c).By contrast, C106-R cells not harbouring KRAS G12C ampli cation did not exhibit the senescence phenotype or markers (Extended Data Figure 4), suggesting that this effect is speci c for KRAS G12C ampli cation.These data suggest that high levels of KRAS signalling, which are needed to drive resistance to the drugs, trigger oncogene-induced senescence upon drug withdrawal.ERK inhibition in these cells by drug treatment is incomplete but re-establishes pathway signalling to a new steady state, similar to active-ERK levels of the parental cells in the absence of drug (Extended Data Figure 1e).
Given the dramatic effect of drug withdrawal in the in vitro model, we wondered if KRAS G12C ampli cation produces a selective disadvantage in patients and monitored the effect of drug withdrawal in three patients with acquired KRAS G12C ampli cation at resistance.In a pancreatic cancer patient, KRAS G12C inhibitor monotherapy was held for eight days for toxicity at the time of emerging clinical resistance.Analysis of ctDNA before and after drug withdrawal revealed a dramatic reduction in both KRAS G12C and KRAS ampli cation one week after disrupting treatment (Figure 4a).A similar result was observed in CRC patients 1 and 5, who both harboured KRAS G12C mutant CRC that had developed multiple resistance alterations, including KRAS G12C ampli cation (Figure 1e, 2a).Again, comparison of ctDNA from before and about 4 weeks after drug withdrawal in each of these patients showed a 2-fold reduction of the signal from KRAS ampli cation.By contrast, the relative frequency of the other preexisting alterations and emergent mutations remained mostly unchanged (Figure 4b, 4c).Together these data show that KRAS G12C ampli cation is a mechanism of secondary resistance that shows tness only in the presence of the selective pressure mediated by drug treatment.
Here we report the genetic mechanisms of secondary resistance to concomitant EGFR and KRAS G12C blockade in KRAS G12C mutant CRC.In agreement with previous studies, our patients show sub-clonal heterogeneity and low variant allele frequency.This may be due in part to the evaluation of ctDNA as this method exposes tumour heterogeneity more than single biopsy specimens 18 .However, our data suggest that low clonal tness may provide the driving pressure for the accumulation of alterations as resistance alterations appear and disappear during continued therapy.Further, our data suggest that a small fraction of cells may be su cient to drive clinical resistance.This concept is in line with recent studies of metastatic behaviour where, for example, single cell RNA sequencing of small cell lung cancers identi ed a rare population of stem-like cells that appears to drive metastatic outcomes in this cancer 19 .
In contrast to the low frequency of other resistance mechanisms, we identify KRAS G12C ampli cation as a recurrent resistance mechanism that tracks with tumour markers and response.KRAS G12C ampli cation engages the senescence program as it results in supra-high levels of ERK signalling when drug is removed.We show that ampli ed KRAS G12C recedes in the absence of drug in patients.Similar to our ndings, BRAF ampli cation upon RAF inhibitor resistance can be modulated by intermittent treatment 20 .Copy number changes may provide a more rapid means to modulate oncogene activity as we nd that KRAS G12C ampli cation emerges faster than resistance mutations in cell lines and that ampli cation carried on extrachromosomal DNA/double minutes, as seen in the RW7213-R cells, may be quickly lost at times of stress 21 .Our data suggest that this drop in the cell free DNA is likely due to the acquisition of a senescent phenotype in cancer cells, therefore intermittent treatment would probably just favour the survival of KRAS G12C ampli ed cells.Moreover, the heterogeneity of resistance mechanisms we and others identi ed suggests that intermittent dosing will not be su cient to overcome resistance.Our data instead raise the potential of novel therapeutic approaches that exploit vulnerabilities due to the senescence program during periods of drug withdrawal to target resistant cells more broadly.SASP chemokines can recruit immune cells suggesting the potential to target resistant cancer cells with immune checkpoint inhibitors [22][23][24][25] .Also, drug withdrawal can potentially be combined with senolytic approaches 26 to facilitate tumour clearance, such as through inhibition of mTOR signaling 27 or antiapoptotic mediators 28 .Moreover, senescence-driven expression of uPAR could also become a target for CART cell therapy 29 .
Further studies will provide important insights on how to effectively target KRAS G12C mutant cancers that developed secondary resistance to KRAS G12C inhibitors as single agents and in combination.
Declarations institution from Genentech, Daiichi Sankyo, Hengrui Therapeutics, Illumina, BioMed Valley Discoveries, AstraZeneca, GRAIL, and Amgen.J.S.R-F. is a member of the scienti c advisory board (with paid honoraria) of Volition Rx, Paige.AI, Invicro, Roche, Genentech, and Ventana, and a consultant with paid fees of Goldman Sachs Merchant Banking.D.B.S. has consulted with and received honoraria from P zer, Loxo Oncology, Lilly Oncology, Vivideon Therapeutics, Q.E.D. Therapeutics and Illumina.N.R. is on the scienti c advisory board (SAB) of Chugai, BeiGene, Fortress Biotech, Daiichi-Sankyo, AstraZeneca, F-Prime, Zai Lab, Arvinas, and Array BioPharma; and he is a past SAB member of Millennium-Takeda, Kadmon, Kura Oncology, and Araxes.N.R. is also a consultant to Novartis Biomed, Boehringer Ingelheim, Tarveda, Foresite Capital, Array BioPharma, and Revolution Medicines; and in recent years has also consulted with Eli Lilly, Merrimack, Kura Oncology, Araxes, and Kadmon.N.R. owns equity in BeiGene, Zai Lab, Fortress Biotech, Kura Oncology, Araxes, Kadmon, and Effector.S.M. has served as a consultant for Boehringer Ingelheim.All other authors declare no potential con icts.

Methods
Cell lines and Compounds RW7213 cell line was cultured in RPMI (Lonza).C106 cell line was cultured in Iscove's modi ed medium (Lonza).Each media was supplemented with 10% FBS, 2mM L-glutamine, 100U/ml penicillin and 100mg/mL streptomycin.Resistant derivatives were grown in media containing cetuximab (50mg/mL) and sotorasib (3mM).C106 cells were purchased from ECACC, RW7213 were provided by Dr. Diego Arando.
All the cell lines were determined to be mycoplasma free using the Venor® GeM Classic kit (Minerva biolabs) and tested by Short Tandem Repeats pro ling at 10 different loci.
Sotorasib and trametinib were purchased from SelleckChem.Cetuximab was purchased from the Pharmacy at Memorial Sloan Kettering Cancer Center.

Cell viability assay
For time course proliferation assays, 1,000 cells were seeded in 96-well culture plates in medium containing 10% FBS.After 24 hours, the indicated concentrations of sotorasib and cetuximab were added to cell in 10% FBS medium (ratio 1:1).Cell viability was assayed with alamarBlue.DMSO-only treated cells were used as control.Assays were performed with eight replicates and were each repeated three times.In all the experiments, plates were incubated at 37°C in 5% CO 2.

Antibodies and Western blotting
After seeding and drug treatments, cells were washed with cold PBS and lysed in RIPA buffer (Pierce #89901) plus phosphatase and protease inhibitors (Thermo Scienti c #1861277, #1861278).Lysates were cleared by centrifugation at 14000rpm at 4°C and quanti ed using BCA method (Pierce #23224).Samples were prepared using LDS+Reducing agent Novex buffers (Invitrogen #NP0008, #NP0009).10 to 20μg of lysates were loaded and run on NuPageTM 4-12% Bis-Tris gels (ThermoFisher #NP0321BOX) followed by transfer to nitrocellulose membranes (Biorad #1620233).Membranes were incubated over night with the indicated antibodies, washed and incubated again for 45 minutes with anti-rabbit or antimouse secondary antibodies.Detection was performed using Immobilion Western (Millipore #WBKLS0500).

RAS-GTP pull-down assay
RAS-GTP pull -down assay was performed according to manufacturer's protocol (Thermo Scienti c #16117).Brie y, 500μg of lysates were loaded into columns together with agarose beads and RAS-RBD bait and incubated for 1 hour at 4°C.After the incubation, beads were washed three times and resuspended in LDS+Reducing agent Novex buffers (Invitrogen #NP0008, #NP0009).A fraction of lysates was used to measure total RAS amount.Pull-Down and total lysates were subjected to western blotting procedure as described above.The kit provided primary antibody against pan-RAS.

SASP Cytokine array
Conditioned media was collected from cells that were cultured in presence or absence of the drug combination.Aliquots of the media were analyzed with a multiplex immunoassay designed for human samples, "Human Cytokine Array / Chemokine Array 48-Plex HD48" (Eve technologies).Cytokine concentration was normalized by cell count.

In vivo studies
The CLR113 PDX was derived from liver metastasis.Tumour tissue was transplanted orthotopically into NSG mice to establish the PDX (IRB protocols 06-107, 14-091).Once a tumour became visible in the rst mouse, it was transplanted and expanded to other animals.Tumour tissue was implanted subcutaneously in the ank of 4-6-week-old NSG female mice and treatment of the mice began when tumour reached approximately 100mm 3 in size.Mice were randomized (n = 5 mice per group) to receive drug treatments or vehicle as control.
Studies were performed in compliance with institutional guidelines under an IACUC approved protocol.The animals were immediately euthanized as soon as the tumours reached the IACUC set limitations.

Patients
All patients were treated on KRAS inhibitor clinical trials approved by MSKCC Institutional Review Board/Privacy Board (protocols 19-408, 20-183).Collection of patient samples were conducted under appropriate Institutional Review Board/Privacy Board protocols and waivers (protocols 06-107, 12-245, 14-019).Participating patients signed written informed consent for these clinical trials and biospecimen protocols.This study was conducted in accordance with ethical guidelines in the Declaration of Helsinki.

Fluorescence In Situ Hybridization (FISH)
FISH analysis was performed on adherent cells.KRAS FISH analysis was performed using a 2-color KRAS/Cen12 probe mix (developed at MSKCC).The probe mix consisted of bacterial arti cial chromosomes (BAC) clones containing the full length KRAS gene (clones RP11-29515 and RP11-707F18; labelled with red dUTP) and a centromeric repeat plasmid for chromosome 12 served as the control (clone pa12H8; labelled with green dUTP).Probe labelling, hybridization, washing, and uorescence detection were performed according to standard procedures.Slides were scanned using a Zeiss Axioplan 2i epi uorescence microscope equipped with a megapixel CCD camera (CV-M4 + CL, JAI) controlled by Isis 5.5.9 imaging software (MetaSystems Group Inc, Waltham, MA).The entire section was scanned through 63X or 100X to assess signal pattern and select representative regions for imaging.Ampli cation was de ned as >10 copies of each locus.

DNA sequencing
Circulating free DNA (cfDNA) Analysis cfDNA analysis performed using the commercially available, targeted next-generation sequencing assays Guardant360 (Guardant Health) (patients 1-5, 11) and ctDx FIRST (Resolution Bioscience) (patients 6-10).Guardant360 CDx is a CLIA-accredited, New York State Department of Health-approved cfDNA assay with analytic and clinical validation previously reported 30,31 .During this study, the assay included assessment of 74-83 genes (depending on panel version ordered) with coverage of single nucleotide variants (SNVs) and select insertions/deletions, ampli cations, and fusions.Resolution Bioscience ctDx FIRST assay includes assessment of 113 genes and detects SNV/Indel Hotspots, SNV/Indel Full CDS, ampli cations, deletions, gene rearrangements, and gene fusions.The ctDx FIRST assay uses a custom bioinformatics pipeline to call variants associated with genomic targets.The longitudinal ctDNA analysis in this study was performed with the Guardant360 CDx assay.

Bulk Tissue Sequencing
Genomic DNA was extracted from cell lines, frozen xenograft tumours, or formalin xed para n embedded (FFPE) patient tissues obtained from biopsies or resections and sequenced with the MSK-IMPACT next-generation sequencing assay 13 .

Single cell DNA sequencing
The C106-R cell line was subjected to single cell sequencing.The cell line was washed with PBS and quanti ed by combining 5uL of cell suspension with an equal amount of Trypan Blue, loaded on chamber slides, and counted with the Countess automated cell counter (Invitrogen).A total of 250,000 cells were used for the barcoding run.In brief, cells were encapsulated with lysis buffer (100 mM Tris at pH 8.0, 0.5% IGEPAL, proteinase K 1.0 mg/ml) in a Tapestri platform (Mission Bio, San Francisco, CA, USA) and further lysed on the thermal cycler with the following conditions: 60 minutes at 50 o C and 10 minutes at 80 o C. The DNA from the encapsulated cell lysate was then primed and barcoded using a custom panel (Mission Bio, San Francisco, CA, USA), which targets hotspot variants of 54 oncogenes and tumour suppressor genes, for a total of 317 amplicons.After exposure to UV light, droplet PCR reactions were thermocycled with the following conditions: products were digested at 37 o C for 60 minutes and posteriorly puri ed using 0.63x of SPRI beads (Beckman Coulter).Sample indices and illumina adapter sequences were then added via a 9 cycle PCR reaction, and a second 0.63x SPRI puri cation was then performed on the completed PCR reactions.Libraries were analysed on a DNA 1000 assay chip with a Bioanalyzer (Agilent Technologies) and sequenced on a NextSeq 550 instrument (Illumina, Inc., San Diego, CA, USA; 150 bp paired-end reads).Sequence data were analysed using the proprietary software provided by Mission Bio 32 .In brief, sequence reads were trimmed for adapter sequences using Cutadapt 33 and mapped to the hg19 human genome using the Burrows-Wheeler Aligner (BWA) 34 after extracting barcode information.Following mapping, on target sequences were selected using standard bioinformatics tool (SAMtools) 35 and barcode sequences were error-corrected based on a white list of known sequences 32 .The number of cells was determined from barcodes based on number of reads assigned to each barcode and amplicon read completeness.GATK 3.7 36 was used to genotype the sample with a joint-calling approach.Genotyping calls were further examined and corrected according to variant allele frequency.For genotype clustering analysis of the two known variants (KRAS G12C and NRAS G12D), cells were included when both variants met the criteria of read depth (³10) and genotyping quality (³60).In addition, subclones having high allele dropout rate (ADO) compared to overall ADO for all cells were further excluded.

Supplementary Files
This is a list of supplementary les associated with this preprint.Click to download.
6 minutes at 98 o C, 10 cycles of 30 seconds at 95 o C, 10 seconds at 72 o C, 9 minutes at 61 o C, and 20 seconds at 72 o C; 10 cycles of 30 seconds at 95 o C, 10 seconds at 72 o C, 9 minutes at 48 o C, and 20 seconds at 72 o C; and a nal step of 2 minutes at 72 o C. PCR